File size: 4,933 Bytes
d2c368c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
90f0f65
 
 
 
 
 
d2c368c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
---
pretty_name: Diogenes Fallout 4 Gameplay and Depth Archive v01
license: other
license_name: original-gameplay-archive-terms
license_link: https://huggingface.co/datasets/DiogenesLab/Diogenes_Gameplay_Fallout_4_Archive_v01/blob/main/NOTICE.md
tags:
- video
- gameplay
- multimodal
- depth
- webdataset
- archival
size_categories:
- n<1K
---

# Diogenes Fallout 4 Gameplay and Depth Archive v01

Public project-closeout archive of **995 clips, 12,120.533 seconds and 363,616
video frames**, from **62 client sessions / 63 source session directories**.
The complete capture batch contains **5,767 original objects / 31,989,102,309 bytes**.
No download application or access code is required. Historical failed QC, missing
modalities and test captures remain visible. No game installers are included.

## Layout and modalities

Tar members use WebDataset-style clip keys such as
`data/session0001-run0001-clip000000`. Each key groups original files:

| Field suffix | Interpretation |
| --- | --- |
| `.video.mp4` | Original H.264 capture; includes gameplay, game menus, loading/tutorial movies, pause and blurred test captures. |
| `.meta.json` | Capture/encoder/game mapping and health: 954 clips use `dm-local-dev-recording-1`, 39 use `dm-local-dev-recording-2`, and 2 use `dm-raw-recording-1`. |
| `.frames.jsonl.zst`, `.events.jsonl.zst` | Zstandard-compressed per-frame input state and input events, including optional depth pairing and camera telemetry. Empty streams are preserved. |
| `.sha256.json` or `.manifest.json` | Original logical-modality SHA-256 maps. |
| `.audio.pcm.zst` | Present for 40 clips. Headerless signed 16-bit little-endian PCM; sample rate/channels in meta, recorded source game-process loopback. Missing audio is not synthesized. |
| `.depth.zip` | Present for 144 clips; original depth-buffer EXRs and `depth-files.json`. |

`metadata/sessionNNNN/` retains session/catalog/mapping profiles, run and heartbeat
state, camera/player pose indexes, depth indexes, probe acknowledgements, frame-pair
and watermark checks, calibration/binding profiles and 20 additional run-level
`depth_00000.zip` bundles. These supporting objects are not additional video clips.
`qc/` contains 33 original acceptance reports. Opaque session/run identifiers and
game-world camera coordinates remain; storage-machine paths are removed from
public member names. Interpret ordering using source metadata/timestamps.

All **164 ZIPs contain 43,822 EXR entries** and their complete `depth-files.json`
indexes (`dm-depth-variant-r-zip-1`). Sampled EXR headers are 1280×720. Depth files
are capture buffers; do not assume metric distance or correct RGB/depth alignment.
Use each run's depth/camera profiles, present sequence, frame pairing and QC.
Repeated depth content is retained (one declared frame digest occurs 1,800 times).
Original representations and failure evidence are preserved rather than corrected.

Twenty copies of the same historical probe DLL were kept in a verified private
archive because embedded build paths expose a developer username. These software
runtime artifacts are not capture payloads. `EXCLUDED_RUNTIME.json` records their
original size/SHA-256 and links to published probe source; no modified DLL is passed
off as an original. All capture media and structured evidence are included.

## Integrity, quality and review

`MANIFEST.jsonl` contains the verified byte size and SHA-256 of all 5,767 public
members, including the twenty run-level ZIPs. The complete reviewed manifest is
transferred. `ARCHIVE_INDEX.json` records each tar hash, size, member count and
fixed verification revision. Pinned Hugging Face LFS checks and anonymous range
downloads passed before the corresponding sources were removed. Original capture
bytes are preserved.

993 clips retain `non_trainable: true` and `production_certified: false`; the two
newer clips omit these flags, and absence is not interpreted as certification.
All 33 acceptance reports failed. No train/test split or training rights are claimed.

All 4,568 outer structured objects (1,340,893 JSONL records) were
fully parsed/scanned, along with every JSON inside each depth ZIP. Empty generic
adapter-ID lists and runtime-local GPU handles are distinguished from personal
identifiers. Representative video frames and MP4 tags were inspected across all
62 client sessions, not every video frame. Audio source provenance is process
loopback; semantic audio review is not claimed.

```python
import json, tarfile
with tarfile.open("downloaded-shard.tar", "r:") as archive:
    for member in archive:
        if member.isfile() and member.name.startswith("data/") and member.name.endswith(".meta.json"):
            m = json.load(archive.extractfile(member))
            print(member.name, m["schema_version"], m["n_frames"])
```

See `NOTICE.md` for rights and eligibility limitations.
Source and schema: https://github.com/CuddlerAI/Diogenes-Archive.